Limitations and Outlook - Scalable Inference for Multi-Target Trackingof Proliferating Cells

problem in Chapter6. Lastly, in Section4.2we proposed a microservices prototype to distribute the computations for pixel classification and thresholding across multiple machines6.

7.2 Limitations and Outlook

The work presented in this thesis has focused on improving the runtime of the Conservation Tracking model for merging and dividing cells, as well as on developing a joint hypotheses selection and tracking pipeline with a method to obtain multiple detection hypotheses through dynamic programming. However, we left the expressivity of the underlying tracking model untouched. Hence there remain several open questions for future research on the model, and other ideas arise from the findings presented above.

• Even though our focus was to develop methods that are general enough to be applied to large variety of problems, biological datasets often have unique features and hence require specialized detection (as in Chapter6) or evaluation strategies. One point we have not touched in this thesis is the resolution of clusters in the model presented in Chapters2

and3. Currently we fit a Gaussian mixture model to each cluster with the number of objects determined by the tracking result, as in [Schiegg et al., 2013]. But by far not all cells, let alone other tracked targets like animals, have an elliptical shape. For optimal performance this step should always be specialized to the scenario at hand.

• The models presented in Chapters2,3and6can only model transitions as pairwise links between a source and a target node. This first order motion model does not allow for in- corporation of velocity or acceleration and is applicable when the cells undergo Brownian motion, but that is not always the case. Some cells, especially bacteria, can self-propel and hence migrate along a certain path. In such cases velocity and acceleration are features that we humans use excessively to re-identify cells in consecutive frames. Integrating such a second or third order motion model would be desirable, but this introduces higher order factors in the graph – ranging over e.g. two transitions and the shared detection node for velocity – which make solving the ILP much harder. One possibility for integrating velocity without adding higher order factors is the model constructed by [Butt and Collins, 2013], which contracts two successive detections plus their transition into one node, hence assignments between those pairings consider velocity. It would be an interesting extension to their model if divisions could also be expressed in a similar way.

• In Chapter6we have used a classifier for transitions, but the Conservation Tracking model from the preceding chapters only used a distance based transition probability. We have performed some experiments with a transition classifier in the Conservation Tracking model as well. For most transitions the classifier score on validation sets was much higher than that of distance-based decisions, because it can also consider size, shape, and color changes. But as the Conservation Tracking model allows clusters of objects, transitions involving the merging or splitting of cell detections look completely different than single object transitions – and hence irritate the classifier. It would be beneficial to develop a more expressive model for transitions than just using distance-based probabilities that supports a varying numbers of objects in source and target detection.

• We showed in Chapter2that the LP relaxation of the network flow formulation of Con- servation Tracking is very tight. Even more so, we know that without division and merger constraints the LP relaxation will yield the optimal integral solution. By initially solving the problem without those additional constraints, and then iteratively adding only the violated constraints in a cutting planes fashion, one might reach the globally optimal solution even faster.

• A lot of tracking mistakes occur when there are segmentation errors, especially missed detections. The common way to deal with missing detection hypotheses is to insert transitions over multiple time steps. Unfortunately, their probability can be hard to tune such that the tracking solution does not prematurely skip available detections. And if the im- portance of these links over multiple frames should be learned by structured learning as in Chapter6, some instances must be annotated as training examples. But as one strives to find a segmentation with as few mistakes as possible, these can be hard to locate, making annotation impractical. Kalman filtering based approaches on the other hand propagate the state of an object into the next frame, allowing it to continue even if the local evidence is not strong. This can bridge the gap if an object becomes e.g. occluded or hard to find due to noise for a short time. By performing a simple Kalman filtering based tracking approach as preprocessing, one could enrich the set of available detections at locations where the segmentation is mistaken.

• Instead of relying on a segmentation to obtain detection hypotheses, we have seen in Section6that a specific model of the target can be required. With the revolutionary performance improvements over the last years in object classification and detection using convolutional neural networks (CNNs) [Krizhevsky et al., 2012,Redmon et al., 2016], real-world object tracking approaches can build on much better sets of detections. While first attempts have been made to use this for cell detection [Xie et al., 2015], there should be a lot of improvement possible through CNNs. In the tracking domain, it would be especially interesting to apply approaches that direct their attention in subsequent frames to a local neighborhood around the previously known position of the target [Ren et al., 2015]. • While the per-frame segmentation can be improved by CNNs [Ronneberger et al., 2015] and state-of-the-art recurrent neural network-based object detectors can be applied to microscopy data followed by tracking-by-detection methods as those presented in this work, completely new tracking methodologies are evolving based on neural networks as well [Alahi et al., 2016, Wang et al., 2015, Nam and Han, 2016]. It is to expect that by a clever application of neural network building blocks combined with a suitable loss function, tracking multiple dividing targets can be approached as well.

• Lastly, evaluating the quality of cell tracking solutions is a difficult task, as metrics do not necessarily capture all aspects of a solution. The Cell Tracking Challenge [Maˇska et al., 2014] used two scores, one for the segmentation quality (average Jaccard score), and one for the edit distance between two tracking graphs consisting of the selected assignments. In Chapter2and3we applied a similar metric for the tracking results, assuming a fixed segmentation. For Chapter6we additionally provide a distance measure of selected hypotheses to the ones present in the ground truth. Nevertheless none of these measures gracefully handles complex types of errors, like a division being detected a frame too early or too late. This is an error that requires changing several links and detections and hence has a big impact on the edit distance, but in terms of biological quality this is far better

7.2. LIMITATIONS AND OUTLOOK

than missing the division completely. [Schiegg et al., 2014] used an evaluation tolerant to that, but still they had to report two numbers, one for the segmentation and one for the tracking quality. It would be desirable to design a measure that represents the overall tracking quality, maybe similar to a Jaccard score of 3D+t volumes.

To conclude, in this work we have considerably improved the scaling behavior of automated tracking methods for dividing targets. We developed a polynomial-time heuristic network flow based solver and a decomposition scheme to parallelize tracking, and distributed the preprocessing workload across multiple machines in a cluster. This, the availability of proof reading tools, and the fact that our heuristic solver is available in ilastik – which frees biologists from the burden of obtaining a license for a commercial ILP solver – enable more biologists around the world to to analyze the ever growing amount of data of proliferating cells in embryogenesis and drug development. While neural network based methods are revamping the world of Com- puter Vision and already began to improve automated tracking, they still require large amounts of training data and dedicated hardware to run efficiently. Furthermore, the images obtained by different microscopy and staining techniques, as well as of varying cell types are so diverse that it seems intractable to obtain good results with pre-trained neural networks across the broad range of recording scenarios. Yet, for only very few of the data there is enough annotated ground truth to train a neural network. As long as this is the case, more general methods will prevail, and hence the approaches developed in this thesis will remain an important contribution.

Bibliography

[Ahuja et al., 1988] Ahuja, R. K., Magnanti, T. L., and Orlin, J. B. (1988). Network flows. Technical report, DTIC Document.

[Alahi et al., 2016] Alahi, A., Goel, K., Ramanathan, V., Robicquet, A., Fei-Fei, L., and Savarese, S. (2016). Social lstm: Human trajectory prediction in crowded spaces. In Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 961–971. [Amat and Keller, 2013] Amat, F. and Keller, P. J. (2013). Towards comprehensive cell lineage

reconstructions in complex organisms using light-sheet microscopy. Development, growth & differentiation, 55(4):563–578.

[Amat et al., 2014] Amat, F., Lemon, W., Mossing, D. P., McDole, K., Wan, Y., Branson, K., Myers, E. W., and Keller, P. J. (2014). Fast, accurate reconstruction of cell lineages from large-scale fluorescence microscopy data. Nature methods.

[Andriluka et al., 2008] Andriluka, M., Roth, S., and Schiele, B. (2008). People-tracking-by- detection and people-detection-by-tracking. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[Andriyenko et al., 2012] Andriyenko, A., Schindler, K., and Roth, S. (2012). Discrete- continuous optimization for multi-target tracking. In Computer Vision and Pattern Recog- nition (CVPR), 2012 IEEE Conference on, pages 1926–1933. IEEE.

[Arteta et al., 2013] Arteta, C., Lempitsky, V., Noble, J. A., and Zisserman, A. (2013). Learning to detect partially overlapping instances. In Proceedings of the IEEE Conference on Computer Vision And Pattern Recognition (CVPR’13), pages 3230–3237, Portland, OR, USA.

[Batra, 2012] Batra, D. (2012). An efficient message-passing algorithm for the m-best map problem. In Proceedings of the 28th Conference on Uncertainty in Artificial Intelligence (UAI 2012).

[Batra et al., 2012] Batra, D., Yadollahpour, P., Guzman-Rivera, A., and Shakhnarovich, G. (2012). Diverse m-best solutions in markov random fields. In Proceedings of the Twelfth European Conference on Computer Vision (ECCV’12), pages 1–16, Florence, Italy.

[Beier et al., 2015] Beier, T., Hamprecht, F. A., and Kappes, J. H. (2015). Fusion moves for cor- relation clustering. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 3507–3516.

[Bellman, 1952] Bellman, R. (1952). On the theory of dynamic programming. Proceedings of the National Academy of Sciences, 38(8):716–719.

[Berclaz et al., 2011] Berclaz, J., Fleuret, F., T¨uretken, E., and Fua, P. (2011). Multiple object tracking using k-shortest paths optimization. Pattern Analysis and Machine Intelligence, IEEE Transactions on, 33(9):1806–1819.

[Berthet and Maizel, 2016] Berthet, B. and Maizel, A. (2016). Light sheet microscopy and live imaging of plants. Journal of microscopy, 263(2):158–164.

[Bertsekas, 1998] Bertsekas, D. P. (1998). Network optimization: continuous and discrete models. Athena Scientific Belmont.

[Bise et al., 2011] Bise, R., Yin, Z., and Kanade, T. (2011). Reliable Cell Tracking by Global Data Association. In IEEE International Symposium on Biomedical Imaging (ISBI), pages 1004–1010.

[Breiman, 2001] Breiman, L. (2001). Random forests. Machine learning, 45(1):5–32.

[Brendel et al., 2011] Brendel, W., Amer, M., and Todorovic, S. (2011). Multiobject tracking as maximum weight independent set. In Computer Vision and Pattern Recognition (CVPR), 2011 IEEE Conference on, pages 1273–1280. IEEE.

[Butt and Collins, 2013] Butt, A. and Collins, R. (2013). Multi-target tracking by lagrangian relaxation to min-cost network flow. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 1846–1853. IEEE.

[Carpenter et al., 2006] Carpenter, A. E., Jones, T. R., Lamprecht, M. R., Clarke, C., Kang, I. H., Friman, O., Guertin, D. A., Chang, J. H., Lindquist, R. A., Moffat, J., et al. (2006). Cell- profiler: image analysis software for identifying and quantifying cell phenotypes. Genome biology, 7(10):R100.

[Castanon and Finn, 2011] Castanon, G. and Finn, L. (2011). Multi-target tracklet stitching through network flows. In IEEE Aerospace Conference, pages 1–7. IEEE.

[Chen et al., 2013] Chen, C., Kolmogorov, V., Zhu, Y., Metaxas, D. N., and Lampert, C. H. (2013). Computing the m most probable modes of a graphical model. In AISTATS, pages 161–169.

[Chen et al., 2014] Chen, C., Liu, H., Metaxas, D., and Zhao, T. (2014). Mode estimation for high dimensional discrete tree graphical models. In Advances in Neural Information Processing Systems (NIPS’14), pages 1323–1331, Montr´eal, Canada.

[Chen et al., 2006] Chen, X., Zhou, X., and Wong, S. T. (2006). Automated segmentation, classification, and tracking of cancer cell nuclei in time-lapse microscopy. IEEE Transactions on Biomedical Engineering, 53(4):762–766.

[Cormen, 2009] Cormen, T. H. (2009). Introduction to algorithms. MIT press.

[Coutu and Schroeder, 2013] Coutu, D. L. and Schroeder, T. (2013). Probing cellular processes by long-term live imaging–historic problems and current solutions. J Cell Sci, 126(17):3805– 3815.

BIBLIOGRAPHY

[Daya et al., 2016] Daya, S., Van Duy, N., Eati, K., Ferreira, C. M., Glozic, D., Gucer, V., Gupta, M., Joshi, S., Lampkin, V., Martins, M., et al. (2016). Microservices from Theory to Practice: Creating Applications in IBM Bluemix Using the Microservices Approach. IBM Redbooks.

[Dijkstra, 1959] Dijkstra, E. (1959). A note on two problems in connexion with graphs. Nu- merische mathematik, 1(1):269–271.

[Dufour et al., 2011] Dufour, A., Thibeaux, R., Labruyere, E., Guillen, N., and Olivo-Marin, J.-C. (2011). 3-d active meshes: fast discrete deformable models for cell tracking in 3-d time-lapse microscopy. IEEE Transactions on Image Processing, 20(7):1925–1937.

[Elowitz et al., 2002] Elowitz, M. B., Levine, A. J., Siggia, E. D., and Swain, P. S. (2002). Stochastic gene expression in a single cell. Science, 297(5584):1183–1186.

[Eppstein, 1998] Eppstein, D. (1998). Finding the k shortest paths. SIAM Journal on Comput- ing, 28(2):652–673.

[Flerova et al., 2012] Flerova, N., Rollon, E., and Dechter, R. (2012). Bucket and mini-bucket schemes for m best solutions over graphical models. In Graph Structures for Knowledge Representation and Reasoning, pages 91–118. Springer.

[Fromer and Globerson, 2009] Fromer, M. and Globerson, A. (2009). An lp view of the m-best map problem. In Advances in Neural Information Processing Systems, pages 567–575. [Fujita et al., 2003] Fujita, Y., Nakamura, Y., and Shiller, Z. (2003). Dual Dijkstra search for

paths with different topologies. In Proceedings of the IEEE International Conference on Robotics and Automation (ICRA’03), volume 3, pages 3359–3364, Taipei, Taiwan.

[Funke et al., 2015] Funke, J., Hamprecht, F. A., and Zhang, C. (2015). Learning to segment: Training hierarchical segmentation under a topological loss. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pages 268–275. Springer. [Haubold et al., 2016a] Haubold, C., Aleˇs, J., Wolf, S., and Hamprecht, F. A. (2016a). A gener-

alized successive shortest paths solver for tracking dividing targets. In European Conference on Computer Vision, pages 566–582. Springer.

[Haubold et al., 2016b] Haubold, C., Schiegg, M., Kreshuk, A., Berg, S., Koethe, U., and Ham- precht, F. A. (2016b). Segmenting and tracking multiple dividing targets using ilastik. In Focus on Bio-Image Informatics, pages 199–229. Springer.

[Held et al., 2010] Held, M., Schmitz, M., Fischer, B., Walter, T., Neumann, B., Olma, M., Peter, M., Ellenberg, J., and Gerlich, D. (2010). Cellcognition: time-resolved phenotype annotation in high-throughput live cell imaging. Nature Methods, 7(9):747–754.

[H¨ockendorf et al., 2012] H¨ockendorf, B., Thumberger, T., and Wittbrodt, J. (2012). Quantita- tive analysis of embryogenesis: a perspective for light sheet microscopy. Developmental cell, 23(6):1111–1120.

[Jaqaman et al., 2008] Jaqaman, K., Loerke, D., Mettlen, M., Kuwata, H., Grinstein, S., Schmid, S. L., and Danuser, G. (2008). Robust single-particle tracking in live-cell time-lapse sequences. Nature methods, 5(8):695–702.

[Jemielita et al., 2013] Jemielita, M., Taormina, M. J., DeLaurier, A., Kimmel, C. B., and Parthasarathy, R. (2013). Comparing phototoxicity during the development of a zebrafish craniofacial bone using confocal and light sheet fluorescence microscopy techniques. Jour- nal of biophotonics, 6(11-12):920–928.

[Joachims et al., 2009] Joachims, T., Finley, T., and Yu, C.-N. J. (2009). Cutting-plane training of structural svms. Machine Learning, 77(1):27–59.

[Jug et al., 2016] Jug, F., Levinkov, E., Blasse, C., Myers, E. W., and Andres, B. (2016). Moral lineage tracing. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 5926–5935.

[Jug et al., 2014] Jug, F., Pietzsch, T., Kainm¨uller, D., Funke, J., Kaiser, M., van Nimwegen, E., Rother, C., and Myers, G. (2014). Optimal joint segmentation and tracking of escherichia coli in the mother machine. In Bayesian and Graphical Models for Biomedical Imaging, pages 25–36. Springer.

[Jüttner et al., ] Jüttner, A., Dezsö, B., and Kovács, P. Lemon: Library for efficient modeling and optimization in networks. Technical report, Dept. of Operations Research, Eötvös Loránd University, Budapest.

[Kausler et al., 2012] Kausler, B. X., Schiegg, M., Andres, B., Lindner, M., Koethe, U., Leitte, H., Wittbrodt, J., Hufnagel, L., and Hamprecht, F. A. (2012). A discrete chain graph model for 3d+ t cell tracking with high misdetection robustness. In European Conference on Computer Vision (ECCV), pages 144–157. Springer.

[Keller, 2013] Keller, P. J. (2013). Imaging morphogenesis: technological advances and biological insights. Science, 340(6137):1234168.

[Keller et al., 2008] Keller, P. J., Schmidt, A. D., Wittbrodt, J., and Stelzer, E. H. (2008). Re- construction of zebrafish early embryonic development by scanned light sheet microscopy. science, 322(5904):1065–1069.

[Kirillov et al., 2015] Kirillov, A., Savchynskyy, B., Schlesinger, D., Vetrov, D., and Rother, C. (2015). Inferring m-best diverse labelings in a single one. In Proceedings of the IEEE International Conference on Computer Vision (ICCV’15), pages 1814–1822, Santiago, Chile. [Koller and Friedman, 2009] Koller, D. and Friedman, N. (2009). Probabilistic graphical mod-

els: principles and techniques. MIT Press, Cambridge, MA, USA.

[Krizhevsky et al., 2012] Krizhevsky, A., Sutskever, I., and Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems, pages 1097–1105.

[Krzic et al., 2012] Krzic, U., Gunther, S., Saunders, T. E., Streichan, S. J., and Hufnagel, L. (2012). Multiview light-sheet microscope for rapid in toto imaging. Nature methods, 9(7):730–733.

[Kschischang et al., 2001] Kschischang, F. R., Frey, B. J., and Loeliger, H.-A. (2001). Fac- tor graphs and the sum-product algorithm. IEEE Transactions on Information Theory, 47(2):498–519.

BIBLIOGRAPHY

[Kuhn, 1955] Kuhn, H. W. (1955). The hungarian method for the assignment problem. Naval research logistics quarterly, 2(1-2):83–97.

[Lampert, 2011] Lampert, C. H. (2011). Maximum margin multi-label structured prediction. In Advances in Neural Information Processing Systems, pages 289–297.

[Lawler, 1972] Lawler, E. (1972). A procedure for computing the k best solutions to discrete optimization problems and its application to the shortest path problem. Management science, 18(7):401–405.

[Lenz et al., 2015] Lenz, P., Geiger, A., and Urtasun, R. (2015). Followme: Efficient online min-cost flow tracking with bounded memory and computation. In IEEE International Con- ference on Computer Vision (ICCV), pages 4364–4372.

[Li et al., 2010] Li, X., Wang, K., Wang, W., and Li, Y. (2010). A multiple object tracking method using kalman filter. In Information and Automation (ICIA), 2010 IEEE International Conference on, pages 1862–1866. IEEE.

[Lou and Hamprecht, 2011] Lou, X. and Hamprecht, F. A. (2011). Structured learning for cell tracking. In Advances in neural information processing systems, pages 1296–1304.

[Magnusson et al., 2014] Magnusson, K., Jalden, J., Gilbert, P., and Blau, H. (2014). Global linking of cell tracks using the viterbi algorithm. Transactions on Medical Imaging, 34(4):911 – 929.

[Magnusson and Jald´en, 2012] Magnusson, K. E. and Jald´en, J. (2012). A batch algorithm using iterative application of the viterbi algorithm to track cells and construct cell lineages. In International Symposium on Biomedical Imaging (ISBI), pages 382–385. IEEE.

[Maˇska et al., 2013] Maˇska, M., Danˇek, O., Garasa, S., Rouzaut, A., Mu˜noz-Barrutia, A., and Ortiz-de Solorzano, C. (2013). Segmentation and shape tracking of whole fluorescent cells based on the chan–vese model. IEEE transactions on medical imaging, 32(6):995–1006. [Maˇska et al., 2014] Maˇska, M., Ulman, V., Svoboda, D., Matula, P., Matula, P., Ederra, C.,

Urbiola, A., Espa˜na, T., Venkatesan, S., Balak, D. M., et al. (2014). A benchmark for com- parison of cell tracking algorithms. Bioinformatics, 30(11):1609–1617.

[McKinney, 2000] McKinney, J. D. (2000). In vivo veritas: the search for tb drug targets goes live. Nature medicine, 6(12):1330–1333.

[Mekterović et al., 2014] Mekterović, I., Mekterović, D., et al. (2014). Bactimas: a platform for processing and analysis of bacterial time-lapse microscopy movies. BMC bioinformatics, 15(1):251.

[Milan et al., 2013] Milan, A., Schindler, K., and Roth, S. (2013). Detection-and trajectory- level exclusion in multiple object tracking. In Proceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition, pages 3682–3689.

[Nam and Han, 2016] Nam, H. and Han, B. (2016). Learning multi-domain convolutional neural networks for visual tracking. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4293–4302.

[Nilsson, 1998] Nilsson, D. (1998). An efficient algorithm for finding the m most probable configurationsin probabilistic expert systems. Statistics and Computing, 8(2):159–173. [Padfield et al., 2011] Padfield, D., Rittscher, J., and Roysam, B. (2011). Coupled minimum-

cost flow cell tracking for high-throughput quantitative analysis. Medical image analysis, 15(4):650–668.

[Pearl, 1988] Pearl, J. (1988). Probabilistic reasoning in intelligent systems: networks of plau- sible inference. Morgan Kaufmann.

[Pirsiavash et al., 2011] Pirsiavash, H., Ramanan, D., and Fowlkes, C. C. (2011). Globally- optimal greedy algorithms for tracking a variable number of objects. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 1201–1208. IEEE.

[Pishchulin et al., 2016] Pishchulin, L., Insafutdinov, E., Tang, S., Andres, B., Andriluka, M., Gehler, P. V., and Schiele, B. (2016). Deepcut: Joint subset partition and labeling for multi person pose estimation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4929–4937.

[Prasad et al., 2014] Prasad, A., Jegelka, S., and Batra, D. (2014). Submodular meets structured: Finding diverse subsets in exponentially-large structured item sets. In Advances in Neural Information Processing Systems (NIPS’14), pages 2645–2653, Montr´eal, Canada. [Rapoport et al., 2011] Rapoport, D. H., Becker, T., Mamlouk, A. M., Schicktanz, S., and

Kruse, C. (2011). A novel validation algorithm allows for automated cell tracking and the extraction of biologically meaningful parameters. PloS one, 6(11):e27315.

[Raviglione and Sulis, 2016] Raviglione, M. and Sulis, G. (2016). Tuberculosis 2015: burden, challenges and strategy for control and elimination. Infectious Disease Reports, 8(2). [Redmon et al., 2016] Redmon, J., Divvala, S., Girshick, R., and Farhadi, A. (2016). You only

look once: Unified, real-time object detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 779–788.

[Ren et al., 2015] Ren, S., He, K., Girshick, R., and Sun, J. (2015). Faster r-cnn: Towards real-time object detection with region proposal networks. In Advances in neural information processing systems, pages 91–99.

[Richardson, 2015] Richardson, C. (2015). Introduction to microservices. URL: https://www.nginx.com/blog/introduction-to-microservices.

[Rollon et al., 2011] Rollon, E., Flerova, N., and Dechter, R. (2011). Inference schemes for m best solutions for soft csps. In Proceedings of the Seventh International Workshop on

In document Scalable Inference for Multi-Target Tracking of Proliferating Cells (Page 89-101)